In the late 1960s, a psychologist named Walter Mischel set up a simple experiment at the Bing Nursery School on Stanford University's campus. A researcher would bring a child, typically between four and six years old, into a small room and seat them at a table. On the table sat a single marshmallow. The researcher explained the terms: you can eat this marshmallow right now, or you can wait until I come back -- about fifteen minutes -- and if the treat is still on the table, you will get two marshmallows instead. Then the researcher left, and a hidden camera recorded what happened next.
Some children ate the marshmallow within seconds of being left alone. Others employed increasingly elaborate strategies to resist: covering their eyes, turning away from the table, singing to themselves, or sitting on their hands. Mischel published the first systematic reports from this work in 1972 in the Journal of Personality and Social Psychology, and follow-up studies in the 1980s suggested that the children who had waited -- who delayed gratification for the larger reward -- showed better outcomes in adolescence on measures including SAT scores, educational attainment, and the ability to cope with frustration. The marshmallow test became one of psychology's most famous demonstrations.
It also became, over subsequent decades, one of its most revised. In 2013, Celeste Kidd, Holly Palmeri, and Richard Aslin published a paper in the journal Cognition showing that the children's decision of whether to wait was powerfully influenced by something the original design had not fully isolated: whether they had reason to trust that the second marshmallow would actually appear. Children who had been exposed, earlier in the same experimental session, to an unreliable researcher who had promised art supplies but failed to deliver them -- who had learned, in other words, that this adult's promises were not worth counting on -- were far less likely to wait for the second marshmallow. They were not demonstrating impulsivity. They were demonstrating rational updating. If your prior experience suggests that deferred rewards have a high probability of not materializing, taking the immediate reward is the sensible strategy.
The marshmallow experiment, in its full complexity, is the perfect window into what researchers call temporal discounting: the universal human tendency to assign less value to rewards that arrive in the future than to rewards available now. The original finding captured something real -- that people differ in their capacity to wait, and that this capacity matters. The Kidd revision captured something equally real and often overlooked: that the rate at which a person discounts the future is not purely an internal trait but is calibrated, partly, by the reliability of the environment. Both truths are embedded in the phenomenon this article examines: hyperbolic discounting, and its behavioral consequence, present bias.
What Hyperbolic Discounting Is
When economists first modeled the way people value future rewards, they adopted what seemed a reasonable simplification: each unit of time adds a constant fractional discount to the value of a future outcome. Wait one more month for a payment, and its present value falls by a fixed percentage. Wait two months, and it falls by that same percentage again. This model -- called exponential discounting -- is mathematically elegant and internally consistent. A person with exponential preferences who prefers $15 today to $20 in one month should also prefer $15 in twelve months to $20 in thirteen months. The relative preference is stable across time. This property is called dynamic consistency.
The problem is that real people do not behave this way.
In a landmark 1981 experiment, economist Richard Thaler asked participants to state the amount they would need to receive in the future to make them indifferent between that amount and a smaller sum available immediately. When the immediate option was $15, participants required roughly $20 to be paid in one month -- a 33 percent premium for a month's wait. So far, consistent with exponential discounting. But when Thaler then asked the same participants how much they would need to receive in thirteen months to be indifferent to $15 in twelve months -- the same one-month delay, shifted one year into the future -- they reported near-indifference between the two options. The same one-month wait that required a $5 premium when it stood between now and next month required almost no premium when it stood between twelve months from now and thirteen months from now.
This is the signature pattern of hyperbolic discounting. The discount applied to any given delay is not constant -- it is much steeper when the delay is near, and much shallower when the delay is distant. The mathematical shape that captures this is a hyperbola rather than an exponential curve. The consequence is preference reversal over time: a person who today prefers $20 in a month to $15 now will, when that month has nearly elapsed, find their preferences flipping -- they will suddenly prefer the $15 available immediately. Plans made in advance appear rational; the moment of execution undoes them.
The Intellectual Lineage
The formal mathematical study of hyperbolic discounting begins with George Ainslie, a psychiatrist and behavioral economist who in 1975 published a paper titled "Specious Reward" in the Psychological Bulletin. Ainslie's central observation was that preference reversals of the kind Thaler would later document experimentally were not anomalies but the predictable output of a discounting function in which the denominator is linear in delay rather than exponential. He argued that this hyperbolic structure was the fundamental cause of what clinicians called impulsiveness, and that it generated an internal conflict between what a person plans (at a distance) and what a person chooses (in the moment). The specious reward of his title was the immediate option that, seen from a distance, looks clearly inferior but, as it approaches, becomes irresistible.
Ainslie's 1975 paper was the theoretical foundation. The model remained primarily in the domain of experimental psychology and behavioral research until 1997, when economist David Laibson at Harvard published "Golden Eggs and Hyperbolic Discounting" in the Quarterly Journal of Economics. Laibson's contribution was to translate the hyperbolic discounting insight into the language of formal economic modeling using what is called the beta-delta framework. In his formulation, an agent discounts all future periods by a common long-run discount factor delta -- the standard exponential component -- but applies an additional immediate discount factor beta specifically to anything that is not the present moment. When beta is less than one, the agent values the present disproportionately relative to all future periods, even those that are close together. The beta-delta model is tractable, embeds in dynamic optimization frameworks, and has become the standard tool for analyzing present bias in economics.
Before Laibson, the economic literature had addressed related problems. Robert Strotz, a University of Illinois economist, published a paper in the Review of Economic Studies in 1956 exploring the implications of non-exponential discounting for the problem of planning and self-control. Strotz recognized that a consumer whose discount function generates preference reversals would be "dynamically inconsistent" -- would systematically deviate from previously made plans -- and that this inconsistency had profound implications for saving, consumption, and commitment. Edmund Phelps and Robert Pollak extended Strotz's framework in a 1968 paper in the Review of Economic Studies, introducing the formal quasi-hyperbolic discount structure in the context of intergenerational utility. Laibson's innovation was to recognize that the same structure applied within an individual's lifetime and to draw out its implications for observable economic behavior.
The behavioral and institutional implications were sharpened considerably by Ted O'Donoghue and Matthew Rabin in their 1999 paper "Doing It Now or Later," published in the American Economic Review. O'Donoghue and Rabin introduced a critical distinction between two types of present-biased agents: the naive agent, who does not recognize his own future tendency toward present bias and therefore makes plans assuming he will behave rationally later; and the sophisticated agent, who correctly anticipates her own present bias and adjusts her behavior accordingly. This distinction proved enormously consequential. Naive agents predictably procrastinate on unpleasant tasks and over-consume in ways they do not foresee; sophisticated agents seek commitment devices to bind their future selves. The same underlying psychological structure generates very different observable behavior depending on whether the agent has accurate insight into his own tendencies.
The Cognitive Science
Two Systems, Two Time Horizons
The neural architecture underlying hyperbolic discounting was mapped in a 2004 paper by Samuel McClure, David Laibson, George Loewenstein, and Jonathan Cohen, published in Science, that stands as one of the most influential brain-imaging studies in behavioral economics. The researchers scanned participants with functional MRI while they made intertemporal choices: options involving immediate or near-immediate monetary rewards versus options involving rewards delayed by weeks or months.
They found that two distinct neural systems were engaged, and that these systems responded selectively to different temporal horizons. Limbic regions -- particularly the ventral striatum and the medial orbitofrontal cortex -- activated preferentially for options that included an immediate reward available today or tomorrow. These are regions associated with the dopaminergic reward system, evolved to respond to concrete, proximate gratifications. The prefrontal cortex and posterior parietal cortex, by contrast, activated for both immediate and delayed options, but showed greater relative engagement when participants ultimately chose delayed rewards.
McClure and colleagues argued that this dissociation mapped onto the beta and delta components of Laibson's model. The limbic beta system responds powerfully to immediately available rewards and generates the present-biased pull toward the here-and-now. The prefrontal delta system represents a more deliberative evaluative process capable of holding future consequences in mind. When the limbic system's activation is strong -- when the immediate reward is salient and proximate -- it tends to dominate. When it is weaker, the prefrontal system's more patient calculations prevail. The result, at the behavioral level, is the steep near-term discount and shallow long-term discount that the hyperbolic function describes.
Subsequent research has refined and complicated this picture. A 2011 meta-analysis by Bartra, McGuire, and Kable, published in NeuroImage, examined over 87 neuroimaging studies of value encoding and confirmed a consistent role for ventral striatum and medial prefrontal cortex in subjective value computation, though the clean beta/delta dissociation has been debated. The deeper point -- that intertemporal choice engages a competition between neural systems with different temporal sensitivities -- has held up well across the subsequent research literature.
The Role of Vividness and Psychological Distance
Temporal discounting is substantially modulated by how concretely a future outcome is imagined. Construal Level Theory, developed by Yaacov Trope and Nira Liberman and synthesized in their 2010 review in Psychological Review, proposes that psychologically distant events -- including those distant in time -- are represented in the mind at a higher level of abstraction than proximate events. You imagine tomorrow's meeting as a specific room, a specific face, a particular conversation; you imagine next year's strategic review as a vague concept. This difference in representational concreteness has direct consequences for how motivating future outcomes feel. Distant rewards are abstract; abstract things lack the emotional vividness that drives the limbic response. Immediate rewards are concrete; concrete things activate the full machinery of desire.
This is why financial advisors who show clients age-progressed photographs of themselves -- their own face as it might look at age seventy -- report higher rates of retirement savings contributions. The experiment was conducted formally by Hal Hershfield and colleagues, published in the Journal of Marketing Research in 2011. Participants who had been exposed to their aged self-image allocated, on average, twice as much of a hypothetical windfall to retirement savings as those who had not. The future self had been made more psychologically real, more concrete, more vivid -- and therefore more capable of competing with the present self for motivational resources.
Comparison of Related Biases
Hyperbolic discounting does not exist in isolation. It overlaps with, generates, and is often confused with several adjacent concepts in behavioral science.
| Concept | Core Definition | Mechanism | Relationship to Hyperbolic Discounting |
|---|---|---|---|
| Hyperbolic Discounting | The discount rate applied to future rewards falls steeply for short delays and shallowly for long delays, generating preference reversals | Nonlinear temporal weighting; disproportionate salience of the immediate | The foundational phenomenon from which present bias follows |
| Present Bias | The tendency to weight costs and benefits occurring in the current moment far more heavily than those occurring in any future period | Beta-delta structure in which all future periods are discounted by an extra factor relative to the present | The behavioral consequence of hyperbolic discounting; sometimes used interchangeably, though present bias can also arise from sources other than the hyperbolic functional form |
| Delay Discounting | Any systematic reduction in the subjective value of rewards as a function of how long one must wait for them | Discounting function of delay; hyperbolic and exponential are both species of delay discounting | Hyperbolic discounting is the empirically observed form of delay discounting; exponential is the normatively consistent form |
| Status Quo Bias | Preference for the current state over alternatives, even when alternatives are objectively superior | Loss aversion applied to the prospect of change; inertia reinforced by omission bias | Related through inertia and the tendency to stick with current behavior; present bias contributes by making the costs of changing immediate and the benefits distant |
| Procrastination | Deferral of tasks, especially unpleasant ones, despite recognizing that deferral has costs | Present-biased preferences make the costs of a task now feel more salient than the future costs of not completing it | One of the most direct behavioral expressions of present bias, particularly in its naive form as theorized by O'Donoghue and Rabin (1999) |
| Projection Bias | Overestimation of the degree to which one's future tastes and preferences will resemble one's current state | Anchoring of future preference estimates to current state; affective forecasting errors | Compounds hyperbolic discounting by making present-oriented choices feel rational; future self is imagined to want what present self wants |
| Impulsivity | The tendency to act on immediate desires with inadequate consideration of delayed consequences | Weak inhibitory control; elevated sensitivity of reward circuitry to proximate stimuli | Present bias is a structural economic description of a preference pattern; impulsivity is a trait-level description of behavior; both describe the same underlying tendency but at different levels of analysis |
What the Research Shows
The empirical literature on hyperbolic discounting and present bias is among the most extensive in behavioral science, spanning laboratory experiments, field studies, neuroimaging, and large-scale economic data.
Thaler's 1981 experiments, conducted at Cornell and later replicated extensively, established the canonical behavioral signature: discount rates that are dramatically higher for short delays than for long ones, and preference reversals that violate the consistency requirement of exponential discounting. Participants in these studies displayed implied annual discount rates of several hundred percent for delays of days or weeks, collapsing to rates closer to 10 to 20 percent per year for delays measured in years -- a pattern no exponential function can fit with a single parameter.
Ainslie and Haendel (1983), in a study published in the International Journal of the Addictions, demonstrated the same hyperbolic preference reversal pattern in populations with substance dependence, suggesting that the heightened present bias associated with addiction was not a categorically different mechanism but an extreme expression of the same underlying discounting structure found in non-clinical populations.
Laibson's 1997 QJE paper used the beta-delta model to explain a striking empirical puzzle in macroeconomics: Americans had, for decades, maintained far lower savings rates than standard life-cycle models predicted they should, given their stated preferences and anticipated retirement needs. The model showed that a moderate degree of present bias -- a beta parameter of around 0.7 to 0.8 -- combined with plausible long-run patience could generate the observed savings patterns. The "golden eggs" of the paper's title referred to the role of illiquid assets -- housing equity, defined-benefit pension claims -- as commitment devices that a present-biased agent exploits to protect her future consumption from her present self's tendency to overconsume.
O'Donoghue and Rabin's 1999 AER paper provided formal results showing that even small degrees of present bias can produce large deviations from optimal behavior, particularly for tasks involving immediate costs and delayed benefits -- the structure of exercise, saving, dieting, and most other self-improvement activities. They also showed that the costs of present bias are larger for naive agents than for sophisticated ones, because naive agents fail to anticipate their own future procrastination and therefore do not seek commitment mechanisms.
Ariely and Wertenbroch, in a 2002 paper published in Psychological Science, ran a field experiment with students in an MIT executive education program. Participants were assigned papers due at the end of a course. Some were allowed to set their own intermediate deadlines -- which would be enforced with grade penalties -- while others had no intermediate deadlines and only a final due date. A third group was given evenly spaced mandatory deadlines by the instructor. The optimal strategy for a student who genuinely believed she would complete work evenly throughout the term would be to take no intermediate deadlines -- they can only reduce flexibility. But students who did impose intermediate deadlines on themselves achieved significantly better performance and grades than those who did not, and the evenly spaced mandatory deadlines produced the best performance of all. Students were, in other words, partially sophisticated about their own present bias -- willing to pay a small cost in flexibility to protect against future procrastination -- but the self-imposed deadlines were less effective than external ones, suggesting imperfect sophistication about how much commitment they actually needed.
A 2015 paper by Shlomo Benartzi and Richard Thaler, published in the Journal of Economic Perspectives, surveyed the evidence for present bias in retirement savings and documented the extraordinary success of the Save More Tomorrow program, in which employees were enrolled in a plan that automatically escalated their retirement savings contribution rate with each future pay raise. Because the commitment to save more was made in advance and took effect only in the future, the immediate-cost structure of present-biased preferences was bypassed. Participation rates and long-run savings balances improved substantially compared to control conditions.
Four Case Studies
Case Study 1: The Retirement Savings Gap
The gap between Americans' stated preferences for retirement saving and their actual behavior is one of the most consequential demonstrations of present bias at scale. Survey data consistently show that Americans believe they should be saving substantially more for retirement than they do. The Vanguard Group's annual "How America Saves" reports have documented, across more than a decade, that median retirement account balances for households near retirement age fall dramatically short of the amounts financial planners recommend -- often by a factor of three or four -- even among households with decades of access to tax-advantaged savings vehicles.
The gap is not primarily explained by income constraints. Households across the income distribution consistently report saving less than they intend to. The mechanism is present bias in its most straightforward form: the cost of saving -- reduced consumption today -- is immediate and concrete, while the benefit -- greater retirement security -- is distant and abstract. A present-biased agent with a beta coefficient of 0.7 effectively discounts retirement consumption relative to present consumption by 30 percent above and beyond her ordinary impatience, meaning that a rational rate of saving from her long-run perspective looks substantially less attractive from her everyday perspective.
The policy solution that exploits this structure -- automatic enrollment with automatic escalation, designed by Thaler and Benartzi as the Save More Tomorrow program and studied formally in a 2004 paper in the Journal of Political Economy -- works precisely because it offloads the immediate cost of saving into the future. Employees commit in advance, when the immediate cost is not yet salient, and the escalation takes effect from future raises rather than present income. The program raised average savings rates from 3.5 percent to 13.6 percent over 40 months at one company where it was implemented. It is now embedded in the Pension Protection Act of 2006 and reaches millions of workers.
Case Study 2: Credit Card Revolving Balances
The credit card industry's business model is, in important part, a business model built around present bias. American consumers carry, in aggregate, more than a trillion dollars in revolving credit card debt -- balances that are not paid off at the end of the billing cycle and that therefore accrue interest, typically at annual rates between 18 and 30 percent. By the standards of any exponential discounting model with plausible parameters, this is deeply irrational: virtually no risk-free or even risk-bearing investment available to ordinary households returns 20 percent annually. A household paying 20 percent annual interest on credit card debt while simultaneously maintaining a savings account earning 3 percent is, from a normative standpoint, making a costly error.
From the perspective of hyperbolic discounting, the behavior is entirely predictable. The purchase is available now. Its pleasure is immediate. The debt repayment is future -- and from the perspective of a present-biased agent making the purchase, the cost of carrying the balance feels discounted to near zero. The same agent who, in January, plans to pay off her December holiday spending by February, finds in February that February's other immediate priorities crowd out the repayment, which shifts forward again. Souleles (1999), analyzing household-level data from the Federal Reserve's Survey of Consumer Finances and published in the Review of Economics and Statistics, found that households frequently held substantial liquid savings while simultaneously carrying high-interest revolving debt -- a pattern only explicable by some form of time-inconsistent preferences or mental accounting.
The credit card industry has systematically exploited this structure. Introductory zero-interest periods are attractive to present-biased consumers precisely because the cost appears to arrive after a delay -- at which point, the beta discount dampens its present value at the time of the purchase decision. Minimum payment structures make debt repayment feel affordable in the present while extending its duration and total cost dramatically. Shui and Ausubel, in a 2005 paper presented at the American Economic Association meetings, conducted an experiment in which they offered credit card consumers different contracts including introductory periods of varying lengths and found that present-biased consumers systematically chose contracts with attractive short-term features that were more costly in the long run.
Case Study 3: The Gym Membership Paradox
Gym memberships are among the most well-documented commercial products in behavioral economics literature, in part because the gap between how people use them and how they should use them is so stark and so measurable. DellaVigna and Malmendier published a paper in the American Economic Review in 2006 analyzing data from three health clubs over three years, covering more than 7,700 memberships.
The central finding was that members who chose flat monthly fee contracts attended, on average, 4.3 times per month -- for an average cost per visit of roughly $17. The same facilities offered ten-visit passes that worked out to approximately $10 per visit. Consumers who were attending four times a month were paying a premium of roughly 70 percent per visit for the monthly contract. Under standard economic assumptions, they should have preferred the cheaper option.
The explanation proposed by DellaVigna and Malmendier, supported by the attrition data, is that consumers were purchasing gym memberships in part as commitment devices and in part due to a specific form of present bias combined with overoptimism: they anticipated going frequently -- their future selves would certainly be motivated and consistent -- but their actual attendance fell far short of projections. They were choosing the monthly contract because their projection of future attendance made it look like the better deal. When actual attendance did not materialize, they continued to maintain the membership -- paying the monthly fee while rarely or never attending -- for an average of 2.3 months longer than optimal. Ending the membership would require a present-moment action; inertia kept them paying.
The gym membership case illustrates a particularly interesting interaction between naive present bias and what the authors call overconfidence about future self-control. Members were not simply impatient. They genuinely believed, at the time of signup, that they would attend frequently. The belief was wrong because they did not accurately anticipate how consistently their future selves would find other immediate activities more attractive than exercise.
Case Study 4: Ulysses Contracts and the Architecture of Commitment
The most direct institutional response to hyperbolic discounting is the commitment device: a mechanism by which a present-biased agent voluntarily restricts her future choices in order to prevent her future self from making decisions she currently anticipates regretting. The concept takes its name from Homer's account of Odysseus binding himself to the mast to hear the Sirens' song without being able to act on the compulsion to follow them into the sea. The sophistication required to design a commitment device is precisely O'Donoghue and Rabin's sophisticated agent -- someone who correctly anticipates her future present bias and takes steps, at a moment of relative patience and clear-headedness, to limit the damage.
In medicine, Ulysses contracts -- legally formalized instructions that patients in lucid periods give for the management of future episodes of impaired capacity, typically in the context of psychiatric illness or addiction -- are the most literal expression of this architecture. A patient in remission may sign an advance directive specifying treatment protocols for a future episode in which, they correctly anticipate, their present-biased preferences for avoiding hospitalization will conflict with their current assessment of what serves their long-run wellbeing.
In personal finance, commitment devices include certificates of deposit with early-withdrawal penalties, illiquid retirement accounts with penalty structures for early disbursement, and automatic savings plans with opt-out friction. Ashraf, Karlan, and Yin published a 2006 study in the Quarterly Journal of Economics testing a commitment savings account product offered to rural bank customers in the Philippines. The account, called SEED (Save, Earn, Enjoy Deposits), paid no higher interest than ordinary accounts but imposed withdrawal restrictions: depositors could only access funds when they had reached a personal savings goal or a target date of their own choosing. Despite the zero interest premium for accepting restrictions, 28 percent of offered customers signed up, and their savings balances were significantly higher after twelve months than those of control customers. People were willing to pay a real cost in liquidity to protect their future savings from their future selves.
The same principle underlies staggered payment structures in behavioral weight-loss programs, the social-commitment components of platforms like stickK.com (where users pledge money to charities they dislike if they fail to meet goals), and the architecture of defined-benefit pension systems that restrict access to accrued benefits until retirement age.
When Present Bias Is Adaptive
The tendency to weight immediate outcomes heavily is not a design flaw in an otherwise well-functioning rational system. It is a design feature of a system calibrated for specific environmental conditions, and those conditions include many situations in which the immediate is, in fact, the appropriate priority.
In environments where future promises are unreliable -- the insight captured by Kidd et al.'s 2013 marshmallow replication -- discounting the future heavily is not a cognitive error but a rational response to the actual reliability of deferred rewards. A child in an unstable household who learns to expect that promises may not be kept is not irrational to discount heavily. An adult in a context where institutions are weak, contracts are poorly enforced, and the future is genuinely uncertain is exercising sound judgment by prioritizing the certain present payoff over the uncertain future one.
In acute emergency conditions, responding to the most proximate threat is the strategy that keeps you alive. The limbic system's bias toward the immediate evolved in precisely such conditions. The organism that stopped to deliberate carefully about the long-run consequences of running from a predator did not survive to reproduce. In environments where immediate threats are real and salient, present-orientation is adaptive.
Even in the domain of modern financial and health decisions, a degree of present weighting serves important functions. Theories of the optimal savings rate under genuine uncertainty about future income, health, and life expectancy cannot determine that the normatively correct discount rate is zero. Uncertainty about whether you will live to collect deferred rewards, about whether institutions will honor their promises, about whether your tastes will remain stable -- all of these provide genuine justification for weighting the near-term more heavily. The problem is not present weighting per se but the excessive steepness of the near-term discount and the preference reversals it generates, which cause behavior to be dynamically inconsistent and systematically to undermine plans that, at a calmer moment, the agent herself endorses.
The limits of the concept also include its empirical measurement challenges. Discount rates measured in laboratory experiments frequently produce implied annual rates that are implausibly high -- hundreds of percent -- which has prompted debate about whether elicitation methods are capturing genuine intertemporal preferences or artifacts of the experimental task, including probability weighting, transaction costs, and distrust of the experimenter's promises. Frederick, Loewenstein, and O'Donoghue's 2002 review in the Journal of Economic Literature surveyed the experimental literature and found that implied discount rates varied enormously across studies -- from negative values to rates exceeding 100,000 percent annually -- a spread they attributed partly to methodological variation and partly to the confounding of multiple psychological processes within any single intertemporal choice.
The Key Insight
What hyperbolic discounting describes is not irrationality in any simple sense but a structural property of human temporal cognition: the present is not just the most recent point on a timeline. It is a qualitatively distinct category. Between now and any future moment there is a discontinuity that no future-to-future comparison contains. This discontinuity generates the characteristic preference reversal -- the person who rationally chooses $20 in thirteen months over $15 in twelve months who, when twelve months have passed, suddenly and urgently wants the $15 -- because the twelve-month option has crossed the threshold from future into now.
The beta-delta model captures this with unusual precision. Long-run patience is real; people do care, genuinely, about their future selves. The extra discount applied to all future periods relative to the present is also real; it is the quantitative expression of the qualitative boundary between the present moment and everything else. Both things are true simultaneously, and their combination generates the systematic self-defeating behavior that O'Donoghue and Rabin characterized in their 1999 framework: intentions that are sincerely formed, promises to future self that are sincerely meant, and violations of both that are predictable in advance if you understand the mechanism.
The practical implication of understanding hyperbolic discounting is not primarily motivational -- knowing about the bias does not reliably dissolve it, for the same reasons that knowing about loss aversion does not dissolve loss aversion. The implication is architectural. If you know that your future self will discount the costs of exercise, eating well, saving, or completing difficult tasks far more heavily than your present self currently believes she will, the effective intervention is not to exhort the future self but to arrange the decision environment now, at a moment of relative clarity, so that the path of least resistance for the future self leads toward the better outcome. Commitment devices, automatic defaults, reduced option availability at critical moments, and social accountability mechanisms are the practical implementations of this insight.
The sophistication that O'Donoghue and Rabin identify as the key variable -- accurate self-knowledge about one's own future present bias -- is, in the end, the thing that separates the person who benefits from understanding hyperbolic discounting from the person who simply adds it to a list of cognitive curiosities. The marshmallow experiment, in its most complete form, tells the same story: it is not purely a test of willpower. It is a test of how accurately a child models the reliability of the future, including the reliability of her own future intentions. Both the environmental insight of Kidd et al. and the individual-difference insight of Mischel converge on the same conclusion: the way you treat time is a function of what you have learned to expect of it.
References
Ainslie, G. (1975). Specious reward: A behavioral theory of impulsiveness and impulse control. Psychological Bulletin, 82(4), 463-496.
Ariely, D., & Wertenbroch, K. (2002). Procrastination, deadlines, and performance: Self-control by precommitment. Psychological Science, 13(3), 219-224.
Ashraf, N., Karlan, D., & Yin, W. (2006). Tying Odysseus to the mast: Evidence from a commitment savings product in the Philippines. Quarterly Journal of Economics, 121(2), 635-672.
Benartzi, S., & Thaler, R. H. (2004). Save More Tomorrow: Using behavioral economics to increase employee saving. Journal of Political Economy, 112(S1), S164-S187.
DellaVigna, S., & Malmendier, U. (2006). Paying not to go to the gym. American Economic Review, 96(3), 694-719.
Frederick, S., Loewenstein, G., & O'Donoghue, T. (2002). Time discounting and time preference: A critical review. Journal of Economic Literature, 40(2), 351-401.
Hershfield, H. E., Goldstein, D. G., Sharpe, W. F., Fox, J., Yeykelis, L., Carstensen, L. L., & Bailenson, J. N. (2011). Increasing saving behavior through age-progressed renderings of the future self. Journal of Marketing Research, 48(SPL), S23-S37.
Kidd, C., Palmeri, H., & Aslin, R. N. (2013). Rational snacking: Young children's decision-making on the marshmallow task is moderated by beliefs about environmental reliability. Cognition, 126(1), 109-114.
Laibson, D. (1997). Golden eggs and hyperbolic discounting. Quarterly Journal of Economics, 112(2), 443-478.
McClure, S. M., Laibson, D. I., Loewenstein, G., & Cohen, J. D. (2004). Separate neural systems value immediate and delayed monetary rewards. Science, 306(5695), 503-507.
O'Donoghue, T., & Rabin, M. (1999). Doing it now or later. American Economic Review, 89(1), 103-124.
Thaler, R. H. (1981). Some empirical evidence on dynamic inconsistency. Economics Letters, 8(3), 201-207.
Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological distance. Psychological Review, 117(2), 440-463.
Frequently Asked Questions
What is hyperbolic discounting?
Hyperbolic discounting is the tendency to discount the value of future rewards at a rate that decreases as the delay increases — placing disproportionate weight on immediate outcomes relative to even slightly delayed ones. Thaler's 1981 experiments showed that people prefer \(15 now over \)20 in a month, but are roughly indifferent between \(15 in a year and \)20 in 13 months. The relative preference reverses as both options move into the future. This time-inconsistency distinguishes hyperbolic from rational (exponential) discounting.
What did the McClure et al. 2004 brain imaging study find?
McClure, Laibson, Loewenstein, and Cohen's 2004 Science paper scanned subjects choosing between immediate and delayed monetary rewards. Choices involving immediate rewards activated the limbic system — including the ventral striatum and medial prefrontal cortex — which showed stronger response for immediate options. Choices involving any delay activated the lateral prefrontal cortex and posterior parietal cortex more uniformly. The two systems competed, with the limbic system's preference for immediacy often overriding the more patient prefrontal evaluation.
What is the gym membership paradox?
DellaVigna and Malmendier's 2006 study found that gym members who paid monthly flat fees of around \(70 attended an average of 4.3 times per month — an effective cost of \)17 per visit. The same gym offered a pay-per-visit option at $10. Members who chose the flat fee paid 70% more per visit than necessary. The authors attributed this to projection bias — members anticipated attending more than they did — combined with present bias: the flat fee feels like a commitment device, but the decision to attend each day is re-made under fresh present bias.
What are commitment devices and do they work?
Commitment devices are mechanisms that constrain future choices to counteract anticipated present bias. Examples include automatic savings plans (removing money before it can be spent), Ulysses contracts in healthcare (advance directives specifying treatment preferences), and deadline structures. Ashraf, Karlan, and Yin's 2006 randomized experiment in the Philippines found that offering a commitment savings account — which restricted withdrawals until a self-set goal date — increased savings rates by 81% relative to control. People who recognize their own present bias can improve outcomes by restricting their future options.
Why do people take out high-interest credit card debt while holding low-interest savings?
Souleles, and Shui and Ausubel's research documented that millions of households simultaneously carry revolving credit card balances at 18-24% interest while holding savings accounts earning 2-4%. Exponential discounting predicts this should not occur — people should use savings to pay down debt. Hyperbolic discounting explains it: the immediate benefit of spending (or the immediate pain of transferring savings) dominates the future cost calculation. The present feels categorically different from even the near future.